OpenAI Rolls Out Enhanced Personalization for ChatGPT

OpenAI is revamping how users interact with ChatGPT, unveiling a refreshed custom instructions menu designed to make personalization more intuitive and accessible.

What’s new:
The updated interface, rolling out now on ChatGPT.com and the Windows desktop app, allows users to tailor the chatbot’s tone and behavior by specifying details like their name, profession, and desired traits such as “Chatty,” “Encouraging,” or “Gen Z.” Mobile and MacOS users will see the update in the coming weeks. However, it won’t be immediately available to users in the EU, Norway, Iceland, Liechtenstein, and Switzerland, per OpenAI.

How it works:
This upgrade builds on ChatGPT’s existing custom instructions, leveraging prompt engineering to guide the AI’s responses based on user input. While the new menu offers a cleaner, more user-friendly design, it doesn’t introduce significant backend changes, relying instead on the same foundational mechanics.

Why it matters:
With over 300 million weekly users, OpenAI is refining ChatGPT to maintain engagement and improve user experience. Recent updates, including live web search and a “Canvas” interface for projects, signal a broader push to keep pace in a competitive AI landscape.

Clarifications:
The updated custom instructions menu is separate from ChatGPT’s memory feature, which retains or discards user-specific details on request. OpenAI’s moderation policies for customizations remain unchanged.

What’s next:
While the rollout is underway, the new interface hasn’t reached all users yet. OpenAI hasn’t commented on the timeline for broader availability, but the enhancements underscore its commitment to more personalized and versatile interactions.

Apple Halts AI News Summaries After High-Profile Errors

Apple is pausing its AI-driven news summarization feature following complaints of false and misleading alerts tied to reputable news outlets.

The AI-generated summaries, part of Apple’s “Apple Intelligence” initiative, inaccurately attributed several false claims to the BBC and other news organizations. Notable errors included a fake BBC alert suggesting Luigi Mangione, accused of killing UnitedHealthcare CEO Brian Thompson, had shot himself. Other fabricated summaries claimed Rafael Nadal had come out as gay and falsely announced a PDC World Darts victory for Luke Littler.

The company plans to suspend the feature in its next software update, stating, “Notification summaries for the news and entertainment category will be temporarily unavailable. We are working on improvements and will make them available in a future software update.” A revised version will include error warnings and italicized text to highlight potential inaccuracies.

The feature, aimed at streamlining news notifications into a single lock-screen alert, has been available in markets like the U.S., UK, Canada, and Australia but excluded from the EU and China. The National Union of Journalists in the UK had previously called for its suspension, citing concerns over misinformation.

AI’s role in content delivery is under scrutiny as high-profile mistakes risk eroding trust in news organizations. The BBC welcomed Apple’s decision, emphasizing the importance of accuracy in maintaining audience trust.

Apple is working to address the flaws while navigating increasing pressure to ensure its AI products don’t contribute to misinformation. The pause signals a cautious step as the company refines its approach to integrating AI into news consumption.

Meta’s Internal Push to Surpass GPT-4 Revealed in Court Filings

Meta’s race to dominate the AI landscape was laid bare in newly unsealed court documents from the ongoing Kadrey v. Meta lawsuit. The filings shed light on the competitive obsession driving the development of Meta’s latest AI model, Llama 3.

Meta’s VP of Generative AI, Ahmad Al-Dahle, underscored the company’s goal to outmatch OpenAI’s GPT-4, referring to it as the benchmark. Internal messages highlight Meta’s aggressive ambitions, including deploying 64,000 GPUs to build “frontier” AI models and asserting dominance over rivals like Anthropic and OpenAI.

While Meta typically releases open models, its leadership compared themselves more closely to closed-model competitors. The French AI startup Mistral, an open competitor, was dismissed by Al-Dahle as “peanuts.”

The filings also detail questionable practices surrounding Llama 3’s training data. Meta executives, including Al-Dahle and researcher Hugo Touvron, discussed improving the dataset mix, even exploring the use of copyrighted materials from the controversial LibGen repository. Prosecutors allege these decisions crossed legal lines in the rush to release competitive models.

Meta CEO Mark Zuckerberg’s internal communications stressed the importance of closing the performance gap with OpenAI and Google. In a 2024 letter, Zuckerberg claimed Llama 3 had become “competitive with the most advanced models,” with plans for future iterations to take the industry lead.

Llama 3, launched in April 2024, succeeded in rivaling closed models from OpenAI, Anthropic, and Google while outperforming open alternatives like Mistral. However, Meta now faces mounting legal scrutiny over its data practices, casting a shadow over its AI advances.

Goldman Sachs CEO Touts AI for Streamlining Financial Tasks

Goldman Sachs CEO David Solomon says AI is transforming investment banking by automating repetitive tasks, dramatically improving efficiency in processes like drafting regulatory filings.

The big picture:
At the Cisco AI Summit in Palo Alto, Solomon revealed that AI can now complete 95% of an SEC S-1 filing—a key document for IPOs—in minutes. Previously, this task required a six-person team and two weeks of effort.

“The last 5% now matters because the rest is now a commodity,” Solomon said, emphasizing the evolving value of human expertise in the age of AI.

Tasks like SEC filings have long been core to investment banks, requiring specialized financial and legal expertise. AI is reshaping this dynamic, forcing banks to rethink how they deliver value to clients.

Goldman Sachs, like its peers, has embraced AI to boost productivity. Solomon noted the importance of giving employees advanced tools to enhance their efficiency and innovate for clients.

By the numbers:
Goldman has maintained a tech-focused workforce, with 11,000 of its 46,000 employees (23%) working as engineers. This reflects a consistent push toward innovation, dating back to at least 2018.

The broader trend:
Wall Street’s embrace of AI extends beyond Goldman Sachs.
• JPMorgan uses AI models to assist portfolio managers and chatbots for routine tasks.
• Morgan Stanley’s “Debrief” tool helps advisors capture meeting notes.
• Wells Fargo leverages AI for customer service.

Between the lines:
The rise of AI has spurred a talent war among banks, with firms like Goldman, Citigroup, and Wells Fargo competing to attract top engineering talent. Executive roles dedicated to AI are also becoming commonplace, fueling rapid innovation across the industry.

As AI tools become standard, the human touch in fine-tuning results will define the competitive edge for banks, reshaping the industry’s future.